# xfilguard > The privacy layer for AI. xfilguard reads what is about to leave, quietly masks what must not (names, numbers, clauses, code) and restores the original for the right reader. Reversible obfuscation at the edge of the model. Key facts about the product: - Reversible obfuscation: sensitive values are substituted with tokens such as `` or `` before a request reaches an AI model; the originals are held in an audit record readable only by the data owner and restored on the way back. - Beyond names and numbers: proprietary formulas, codenames, pricing tables, and unreleased strategy are protected the same way as classic PII (names, SSNs, emails). - Works in 60+ languages, including multilingual documents, code-switching, and regional variants; values are restored in the same script they were written in. - Handles every format: plain text, docx, pdf, csv, and audio, including messy real-world inputs like OCR artifacts, handwritten forms, and ASR transcripts. - Synthetic substitutes: when a value must be replaced rather than hidden, xfilguard can produce a plausible stand-in that preserves statistical shape without leaking the source. - How it works, in three steps: (1) Read: intercept the outbound request and scan it for what should not leave; (2) Mask: collapse sensitive values into tokens while preserving the structure of the request; (3) Restore: put the originals back for the right reader when the model responds. - Model-agnostic: works ahead of any provider, so changing model vendor does not change the policy. - Not a firewall and not a censor: the request goes through with the secrets removed, rather than being blocked and queued for approval. - Positioning: a privacy layer, a translator, reversible, model-agnostic, and quiet. Not a firewall, not a censor, not destructive, not tied to one vendor. - Compliance: every mask leaves an audit record (timestamp, policy, model, entity type, verb, retention window), readable only by the data owner. Attested regulations: GDPR (EU), HIPAA (US), CCPA/CPRA (California), SOC 2 Type II (global), PCI-DSS (global), APPI (Japan). - Category terms people use for this: PII masking, prompt redaction, de-identification, data loss prevention (DLP) for LLMs, AI privacy layer, prompt security. - Contact: hello@xfilguard.com, or book a demo through the site. ## Docs - [Landing page in Markdown](https://xfilguard.com/index.html.md): the complete site content (what xfilguard does, how it works, where it sits, how it complies, and the FAQ) as clean markdown for agents. - [Everything in one file](https://xfilguard.com/llms-full.txt): the same content as a single plain-text document. ## Optional - [Landing page (HTML)](https://xfilguard.com/): prerendered to static HTML, so the full content is present without executing JavaScript.